Are there? Are there any large-scale regularities explaining how cultural items become “fashionable”? This question was asked in Alfred Kroeber’s paper “On the principle of order in civilization as exemplified by changes of fashion” precisely 100 years ago, in 1919. To answer it, Kroeber made something rather unexpected: a quantitative study of… dresses. The man of science, in his own words, “went so far as to turn the leaves of volume after volume of a Parisian journal devoted to dress”. With a ruler, of course. Kroeber measured the lengths of skirts, their diameter, and even the depth of decoultage. His discovery? In fact, not a big one: several dissimilar trends without explanation.
Flashforward a century: gigabytes of the cultural “big data” are piling up in various corners of the Internet. So, we must have a better answer, don’t we? Several months ago, a group of complexity scientists — Peter Klimek, Robert Kreuzbauer, and Stefan Thurner — published an interesting paper: a study of fashions in music industry. They performed their study with what Kroeber didn’t have: a large database. Almost 8 million music albums, released from 1952 to 2015. Armed with this information, the paper tests three hypotheses of how fashion works — three hypothetical “laws”.
What are these hypotheses? They are summarized on Figure 1, adapted from the study.
First, the costly signalling theory of fashion. According to it, prestigious individuals always seek ways of distinguishing themselves from others (from poor proletarians) — by clearly signalling their status. And what can be clearer than owning something no one else can afford? Owning “costly” items. To do their job, however, such items should be distinctive. (That’s the reason why clothes from Dolce & Gabbana have their brand name written in huge letters on everything, while clothes from H&M don’t.) But being distinctive isn’t easy: proletarians always seek ways of mimicking costly signals. And so, those who actually have money and status, are pushed to invent new ways of showing off. This is a never-ending chase, and the whole industry of knockoffs owes its existence to it.
Here’s a recent example: in 2017, Apple presented iPhone X, a rather “costly” device — and a distinctive one. It could be immediately recognised due to its “notch”: a cutout in the top part of the screen hosting a selfie camera and various sensors for the iPhone’s novel face-ID technology.
Very soon the notch became a status symbol. So, it took other phone manufactuters less than a year to develop their (much cheaper) phones with a similar notch (Figure 2). Funnily, for most of them the notch wasn’t really functional: none of them had the complex sensors for face-ID. It simply was an attempt to mimic the prestigious status symbol. Thus, Apple had to invent new ways to clearly signal the high status of their smartphones.
Second theory: the random pattern theory of fashion — more commonly known as cultural drift. According to it, people copy each other in a random way: we simply mimic whatever we see around us — no matter high status or not. A classic example here is baby names. Names rarely harm or help their bearers: as a result, there is virtually no “natural selection” of names. They simply drift, resulting in rather random naming fashions. As usual, there are interesting exceptions. For example, in today’s Germany there are almost no babies named Adolf (surprise-surprise!) Being named Adolph can be harmful. And so — the great extinction of Adolfs (Figure 3). But this case of non-random evolution is an exception; the rule, in the history of names, seems to be random copying, drift.
Finally, a theory suggested by the authors of the paper: the counter-dominance theory of fashion. According to it, whenever a cultural item — say, a genre of music — becomes popular, it doesn’t start “running away” from the copycats, like the elites in the costly signalling theory; neither is it copied randomly. Instead, new fashions emerge in response to the dominant fashion: they obtain features dissimilar from those of the dominating “elite”.
In the paper, these three theories are tested on the case of music history. For that, the authors use Discogs, a large crowdsourced database that stores information about the release of music albums. Each album is assigned one or several genre tags (“swing”, “disco”, “techno”…). Also, the data contains information about the musical instruments used in each particular album.
First, music genres. Figure 4 shows the popularity of each genre (based on the number of albums using each genre tag). Clearly, we see waves, or cycles, of popularity. The initial wave of “vocal” music gradually declines in popularity, and instead another wave rises: “rock’n’roll”. Which, in turn, is later surpassed by a new genre: “pop rock”. Ad infinitum.
Waves of fashion. The question is: what is the driving force behind them? Costly signalling? Random drift? Counter-dominance?
Klimek, Kreuzbauer, and Thurner do find that new styles have dissimilar sets of instruments used. Besides, they build three mathematical models, for each of three main theories of fashion, and test which one is best at explaining the cyclical pattern on Figure 4. The result: counter-dominance theory, suggested by the authors in this paper, is better than others. The authors conclude:
Here, we provide an entirely data-driven and quantitative answer to this question [whether there are specific patterns in the social dynamics that underlie art and fashion cycles and — if yes — what the mechanisms driving them are]: cyclical changes of complex mass-cultural symbol systems manifested in art and fashion styles are driven by outside groups that successfully challenge the current elites.
Now, does this mean that other theories — costly signalling and drift — are wrong? The authors clearly oppose their theory to the other two: three theories compete with each other, and the counter-dominance theory wins in this competition. But how general is this theory? Does it explain “fashion” in any domain of culture?
The problem is, “fashion” is a vague notion. Earlier, I illustrated each theory with an example — to show how different these examples are; and each theory reflects these differences. The costly signalling theory is a theory of status. Thus, it can be applied to the symbols of status: expensive watches, supercars, or iPhones. But of course it will fail if applied to anything else. Is music a status symbol? Maybe — in some rare cases (attending opera may indeed signal some social status, or at least used to in the past) — but rock’n’roll or techno certainly aren’t.
The same can be said about random drift. It is easier to imagine random copying of cultural items that do not have strong advantages or disadvantages — like names. But it can hardly be said about music. Surely, randomness plays a big role in the success of individual hit songs, but, on average, listeners do prefer songs that they consider better.
So, are there laws of fashion? Probably, this question could be improved. “Fashion” is a cumbersome, bulky notion that can relate to various mechanisms in various cultural domains. “Fashion” in music is not the same as “fashion” in dresses. And “fashion” in dresses is not the same as “fashion” in baby names. Maybe, instead of discussing fashion as a singular phenomenon, we should shift to discussing fashions — in plural. Different fashions, with different mechanisms, or “laws”.
Are we closer to explaining fashions than Kroeber, a century ago? Definitely yes — due to studies like this one. But there still is a long way to go. And if, on this way, we drop the singular notion of fashion at all — this will probably be for good.